[DISCUSS] Shared NFS Zone-wide (primary) Block Storage

[Hari Kannan <ha...@citrix.com>]

Hello All,

I wish to propose Shared NFS Zone-wide (primary) Block Storage CloudStack - I have added some details here (https://cwiki.apache.org/confluence/display/CLOUDSTACK/Shared+NFS+Zone-wide+(primary)+Block+Storage)  along with a JIRA ticket 724

Please review and comment

Hari Kannan

Re: [DISCUSS] Shared NFS Zone-wide (primary) Block Storage

[Marc Cirauqui <mc...@gmail.com>]

Hi,

That would be nice, but, just a thought, why just NFS? Why not iSCSI o
"PreSetupt" too?

thx


On Thu, Jan 3, 2013 at 12:27 AM, Hari Kannan <ha...@citrix.com> wrote:

>
> Hello All,
>
> I wish to propose Shared NFS Zone-wide (primary) Block Storage CloudStack
> - I have added some details here (
> https://cwiki.apache.org/confluence/display/CLOUDSTACK/Shared+NFS+Zone-wide+(primary)+Block+Storage)
>  along with a JIRA ticket 724
>
> Please review and comment
>
> Hari Kannan
>

Re: [DISCUSS] Shared NFS Zone-wide (primary) Block Storage

[Wido den Hollander <wi...@widodh.nl>]

On 01/03/2013 12:27 AM, Hari Kannan wrote:
>
> Hello All,
>
> I wish to propose Shared NFS Zone-wide (primary) Block Storage CloudStack - I have added some details here (https://cwiki.apache.org/confluence/display/CLOUDSTACK/Shared+NFS+Zone-wide+(primary)+Block+Storage)  along with a JIRA ticket 724
>
> Please review and comment

I've been thinking about this and IF we do this, shouldn't we make sure 
all primary storages are supported?

I'm also looking for a valid reason for doing this. Why not expand your 
cluster size?

Wouldn't doing this break the whole idea of zones, pods and clusters? 
Since they all share the same resources you can have multiple clusters 
going down (or even your whole zone) by just one component failing.

I know this is the responsibility of the guy who builds the CloudStack 
cluster, but still, why have more clusters when they share the same 
primary storage? You just add one SPoF spread out over multiple clusters.

Wido

>
> Hari Kannan
>

Re: [DISCUSS] Shared NFS Zone-wide (primary) Block Storage

[Marcus Sorensen <sh...@gmail.com>]

Sure, that makes sense, Alex. If someone wants to create a zone-wide
primary storage for data volumes, then that data volume can be migrated
between VMs in separate clusters.


On Thu, Jan 3, 2013 at 2:06 PM, Marcus Sorensen <sh...@gmail.com> wrote:

> Clusters still work to segment which compute the VMs run on, and can do
> tags or other networking related segmentation.
>
> I can understand this for something like RBD, you could have all of your
> clusters backed by the same Ceph cluster. But even then I think it makes
> more sense to define primary storage per-cluster, even if you define the
> same primary storage for multiple clusters. Unless we're just saying we
> want the ability to do away with clusters altogether and just have a giant
> pool of compute and storage. If people want that they can do it with a
> single, large cluster. Ultimately, even if I were using the same primary
> storage SAN, distributed object store, etc for multiple clusters, it scales
> better to partition them into separate buckets/diskarrays/whatever.
>
> I don't know, I can see what it's trying to accomplish, but I'm not
> necessarily sold on the utility or that it can't already be done in a
> reasonable fashion without breaking the primarypool/cluster relationship.
>
> On Thu, Jan 3, 2013 at 1:54 PM, Wido den Hollander <wi...@widodh.nl> wrote:
>
>> Isn't a cluster supposed to be an isolated "island" of machines which do
>> not have any ties with other clusters other then being in the same pod/zone?
>
>
>

Re: [DISCUSS] Shared NFS Zone-wide (primary) Block Storage

[Marcus Sorensen <sh...@gmail.com>]

Clusters still work to segment which compute the VMs run on, and can do
tags or other networking related segmentation.

I can understand this for something like RBD, you could have all of your
clusters backed by the same Ceph cluster. But even then I think it makes
more sense to define primary storage per-cluster, even if you define the
same primary storage for multiple clusters. Unless we're just saying we
want the ability to do away with clusters altogether and just have a giant
pool of compute and storage. If people want that they can do it with a
single, large cluster. Ultimately, even if I were using the same primary
storage SAN, distributed object store, etc for multiple clusters, it scales
better to partition them into separate buckets/diskarrays/whatever.

I don't know, I can see what it's trying to accomplish, but I'm not
necessarily sold on the utility or that it can't already be done in a
reasonable fashion without breaking the primarypool/cluster relationship.

On Thu, Jan 3, 2013 at 1:54 PM, Wido den Hollander <wi...@widodh.nl> wrote:

> Isn't a cluster supposed to be an isolated "island" of machines which do
> not have any ties with other clusters other then being in the same pod/zone?

RE: [DISCUSS] Shared NFS Zone-wide (primary) Block Storage

[Alex Huang <Al...@citrix.com>]

> -----Original Message-----
> From: Wido den Hollander [mailto:wido@widodh.nl]
> Sent: Thursday, January 03, 2013 12:54 PM
> To: cloudstack-dev@incubator.apache.org
> Subject: Re: [DISCUSS] Shared NFS Zone-wide (primary) Block Storage
> 
> 
> 
> On 01/03/2013 07:04 PM, Edison Su wrote:
> >
> >
> >> -----Original Message-----
> >> From: David Nalley [mailto:david@gnsa.us]
> >> Sent: Thursday, January 03, 2013 7:31 AM
> >> To: cloudstack-dev@incubator.apache.org
> >> Subject: Re: [DISCUSS] Shared NFS Zone-wide (primary) Block Storage
> >>
> >> On Wed, Jan 2, 2013 at 6:27 PM, Hari Kannan <ha...@citrix.com>
> >> wrote:
> >>>
> >>> Hello All,
> >>>
> >>> I wish to propose Shared NFS Zone-wide (primary) Block Storage
> >>> CloudStack - I have added some details here
> >>>
> >>
> (https://cwiki.apache.org/confluence/display/CLOUDSTACK/Shared+NFS+Zo
> >> n
> >>> e-wide+(primary)+Block+Storage)  along with a JIRA ticket 724
> >>>
> >>> Please review and comment
> >>>
> >>> Hari Kannan
> >>
> >> So I have to say that I don't really understand the reason for wanting to
> >> invest the effort in doing this, and fear it will just result in a bad
> experience
> >> for the majority of folks who embark upon it.
> >> For a small setup this will work fine, but that small setup is also likely to
> only
> >> have a few clusters. I fear that in all but the most niche cases that this
> simply
> >> doesn't scale.
> >
> > It's not about shared nfs zone-wide primary storage, it is about: can we
> support zone-wide primary storage in cloudstack? Don't matter what kind of
> primary storage people want to use, it can be nfs(if they want, for whatever
> reason), or solidfire etc.
> > Currently, cloudstack just can't do that.
> >
> 
> Ok, clear :-)
> 
> But still, what would then be the purpose of having multiple clusters
> when they all share the same primary storage?
> 
> If that primary storage fails all your clusters go down with it.
> 
> Or am I thinking in the wrong way?
> 
> Isn't a cluster supposed to be an isolated "island" of machines which do
> not have any ties with other clusters other then being in the same pod/zone?
> 
> Wido
> 

Hey Wido,

Welcome back!

We've seen a cluster's management layer went bad but the storage did not.  I think if you look at this feature as every primary storage is zone wide, your concern is justified.  However, if you look at it as someone wants to be able to do root disk on clustered-based primary storage but ability to move data disk around to different vms, this is makes a lot of sense to implement.  I echo Edison's message.  It's about the enabling the capability and let others to worry about usage and scaling. 

--Alex

Re: [DISCUSS] Shared NFS Zone-wide (primary) Block Storage

[Wido den Hollander <wi...@widodh.nl>]

On 01/03/2013 07:04 PM, Edison Su wrote:
>
>
>> -----Original Message-----
>> From: David Nalley [mailto:david@gnsa.us]
>> Sent: Thursday, January 03, 2013 7:31 AM
>> To: cloudstack-dev@incubator.apache.org
>> Subject: Re: [DISCUSS] Shared NFS Zone-wide (primary) Block Storage
>>
>> On Wed, Jan 2, 2013 at 6:27 PM, Hari Kannan <ha...@citrix.com>
>> wrote:
>>>
>>> Hello All,
>>>
>>> I wish to propose Shared NFS Zone-wide (primary) Block Storage
>>> CloudStack - I have added some details here
>>>
>> (https://cwiki.apache.org/confluence/display/CLOUDSTACK/Shared+NFS+Zo
>> n
>>> e-wide+(primary)+Block+Storage)  along with a JIRA ticket 724
>>>
>>> Please review and comment
>>>
>>> Hari Kannan
>>
>> So I have to say that I don't really understand the reason for wanting to
>> invest the effort in doing this, and fear it will just result in a bad experience
>> for the majority of folks who embark upon it.
>> For a small setup this will work fine, but that small setup is also likely to only
>> have a few clusters. I fear that in all but the most niche cases that this simply
>> doesn't scale.
>
> It's not about shared nfs zone-wide primary storage, it is about: can we support zone-wide primary storage in cloudstack? Don't matter what kind of primary storage people want to use, it can be nfs(if they want, for whatever reason), or solidfire etc.
> Currently, cloudstack just can't do that.
>

Ok, clear :-)

But still, what would then be the purpose of having multiple clusters 
when they all share the same primary storage?

If that primary storage fails all your clusters go down with it.

Or am I thinking in the wrong way?

Isn't a cluster supposed to be an isolated "island" of machines which do 
not have any ties with other clusters other then being in the same pod/zone?

Wido

>>
>> Also - trying to emulate Amazon's EBS with something like NFS that is unlikely
>> to scale to EBS proportions seems an interesting choice. I could understand
>> using one of the distributed filesystems like Gluster or Ceph to do this, so
>> why NFS?

RE: [DISCUSS] Shared NFS Zone-wide (primary) Block Storage

[Edison Su <Ed...@citrix.com>]

> -----Original Message-----
> From: David Nalley [mailto:david@gnsa.us]
> Sent: Thursday, January 03, 2013 7:31 AM
> To: cloudstack-dev@incubator.apache.org
> Subject: Re: [DISCUSS] Shared NFS Zone-wide (primary) Block Storage
> 
> On Wed, Jan 2, 2013 at 6:27 PM, Hari Kannan <ha...@citrix.com>
> wrote:
> >
> > Hello All,
> >
> > I wish to propose Shared NFS Zone-wide (primary) Block Storage
> > CloudStack - I have added some details here
> >
> (https://cwiki.apache.org/confluence/display/CLOUDSTACK/Shared+NFS+Zo
> n
> > e-wide+(primary)+Block+Storage)  along with a JIRA ticket 724
> >
> > Please review and comment
> >
> > Hari Kannan
> 
> So I have to say that I don't really understand the reason for wanting to
> invest the effort in doing this, and fear it will just result in a bad experience
> for the majority of folks who embark upon it.
> For a small setup this will work fine, but that small setup is also likely to only
> have a few clusters. I fear that in all but the most niche cases that this simply
> doesn't scale.

It's not about shared nfs zone-wide primary storage, it is about: can we support zone-wide primary storage in cloudstack? Don't matter what kind of primary storage people want to use, it can be nfs(if they want, for whatever reason), or solidfire etc.
Currently, cloudstack just can't do that.

> 
> Also - trying to emulate Amazon's EBS with something like NFS that is unlikely
> to scale to EBS proportions seems an interesting choice. I could understand
> using one of the distributed filesystems like Gluster or Ceph to do this, so
> why NFS?

Re: [DISCUSS] Shared NFS Zone-wide (primary) Block Storage

[David Nalley <da...@gnsa.us>]

On Wed, Jan 2, 2013 at 6:27 PM, Hari Kannan <ha...@citrix.com> wrote:
>
> Hello All,
>
> I wish to propose Shared NFS Zone-wide (primary) Block Storage CloudStack - I have added some details here (https://cwiki.apache.org/confluence/display/CLOUDSTACK/Shared+NFS+Zone-wide+(primary)+Block+Storage)  along with a JIRA ticket 724
>
> Please review and comment
>
> Hari Kannan

So I have to say that I don't really understand the reason for wanting
to invest the effort in doing this, and fear it will just result in a
bad experience for the majority of folks who embark upon it.
For a small setup this will work fine, but that small setup is also
likely to only have a few clusters. I fear that in all but the most
niche cases that this simply doesn't scale.

Also - trying to emulate Amazon's EBS with something like NFS that is
unlikely to scale to EBS proportions seems an interesting choice. I
could understand using one of the distributed filesystems like Gluster
or Ceph to do this, so why NFS?

RE: [DISCUSS] Shared NFS Zone-wide (primary) Block Storage

[Alex Huang <Al...@citrix.com>]

> -----Original Message-----
> From: Chip Childers [mailto:chip.childers@sungard.com]
> Sent: Thursday, January 03, 2013 1:21 PM
> To: cloudstack-dev@incubator.apache.org
> Subject: Re: [DISCUSS] Shared NFS Zone-wide (primary) Block Storage
> 
> On Wed, Jan 2, 2013 at 6:27 PM, Hari Kannan <ha...@citrix.com>
> wrote:
> >
> > Hello All,
> >
> > I wish to propose Shared NFS Zone-wide (primary) Block Storage
> CloudStack - I have added some details here
> (https://cwiki.apache.org/confluence/display/CLOUDSTACK/Shared+NFS+Zo
> ne-wide+(primary)+Block+Storage)  along with a JIRA ticket 724
> >
> > Please review and comment
> >
> > Hari Kannan
> >
> 
> I'm +1 to being able to create a zone-wide primary storage target, but
> I'd like to see the feature name changed to be simply that (and not
> talk about NFS). ;-)

+1 It should support all.  What we can do is say in the feature which technology may happen first.

--Alex

Re: [DISCUSS] Shared NFS Zone-wide (primary) Block Storage

[Chip Childers <ch...@sungard.com>]

On Wed, Jan 2, 2013 at 6:27 PM, Hari Kannan <ha...@citrix.com> wrote:
>
> Hello All,
>
> I wish to propose Shared NFS Zone-wide (primary) Block Storage CloudStack - I have added some details here (https://cwiki.apache.org/confluence/display/CLOUDSTACK/Shared+NFS+Zone-wide+(primary)+Block+Storage)  along with a JIRA ticket 724
>
> Please review and comment
>
> Hari Kannan
>

I'm +1 to being able to create a zone-wide primary storage target, but
I'd like to see the feature name changed to be simply that (and not
talk about NFS). ;-)